Automated Prediction of Malignant Melanoma using Two-Stage Convolutional Neural Network.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin lesions can be benign or malignant, with malignant lesions indicating skin cancer.
• Skin cancer is the most common cancer in the US, and traditional detection methods can be painful.
• Automated detection systems offer a promising alternative for early and accurate diagnosis.

Purpose of the Study:

• To develop and evaluate an automated system for skin cancer prediction and classification.
• To investigate the efficacy of a two-stage Convolutional Neural Network (CNN) approach.
• To compare the performance of feature fusion with traditional methods.

Main Methods:

• A two-stage CNN was employed for automated skin cancer prediction.
• The first CNN stage extracted low-level features, and the second extracted high-level features.
• Features were fused with ABCD (Asymmetry, Border irregularity, Colour variation, Diameter) technique and classified using ensemble and machine learning models.

Main Results:

• The first stage CNN achieved 97.92% accuracy in dataset creation.
• The second stage CNN achieved 98.86% accuracy in feature selection.
• Feature fusion in the two-stage model demonstrated superior classification performance.

Conclusions:

• The proposed two-stage CNN model with feature fusion significantly enhances skin cancer prediction accuracy.
• This automated approach offers a potentially less invasive and more efficient method for skin cancer diagnosis.
• Further validation on diverse datasets is recommended to confirm generalizability.